Layer 1 of 8 · 8 companies
AI clouds: the buyers
Can the biggest buyers keep paying for this out of their own cash?
The five largest buyers spent $566B on capital projects in their latest four quarters, and the GPU-only clouds borrow to spend even faster relative to their size. This layer’s spending is every other layer’s revenue, so the question for the whole stack starts here: can the buyers keep paying?
- capex by Microsoft, Amazon, Alphabet, Meta and Oracle, latest four quarters
- $566B
- capex by Microsoft, Amazon, Alphabet, Meta and Oracle, latest four quarters
- of their operating cash flow spent on capex in Q2 2026
- 97%
- of their operating cash flow spent on capex in Q2 2026
- of data-center leases signed but not yet started, outside capex
- $1.08T
- of data-center leases signed but not yet started, outside capex
- Google Cloud revenue growth over its latest four quarters (AWS +28%)
- +76%
- Google Cloud revenue growth over its latest four quarters (AWS +28%)
Five companies, one decision
The AI build-out is, financially, the decision of a handful of companies. Microsoft, Amazon, Alphabet, Meta and Oracle reported $566B of capital spending over their latest four quarters, up from $184B two years earlier. Not all of it is for AI, and none of them splits the figure out, but their filings attribute the growth to AI and cloud infrastructure, and the suppliers’ revenue confirms where it went.
Treat these companies as two kinds of buyer. Microsoft, Amazon, Alphabet and Oracle rent capacity to others, so their spending is a bet that customers will pay for it through their cloud businesses. Meta builds for its own apps, so its return shows up in advertising: better ranking, better ad targeting and the products it hopes to build on its own models. The first kind can point to cloud revenue as the return. The second has to argue it from engagement and ad pricing, which is harder to see from outside.
Together they earn $1.91T of revenue, +35% in two years, at an operating margin of 27%. That revenue is mostly not AI: it includes Amazon’s shop, Alphabet’s search ads, Microsoft’s Office and Windows. The point of listing it is scale. These companies can afford the spending; the question is how long they will want to, and what it does to their returns on capital.
Quarterly revenue: AI clouds
The layer’s 7 US-listed companies, five largest shown separately; Other is CoreWeave, Cloudflare
- Amazon
- Alphabet
- Microsoft
- Meta
- Oracle
- Other
Source: company 10-K and 10-Q filings (SEC XBRL); fiscal quarters aligned to the nearest calendar quarter. Companies appear from their first reported quarter, so newly listed ones (such as CoreWeave, Sandisk or Astera Labs) add to the total partway through.
Show the numbers
| Quarter | Amazon | Alphabet | Microsoft | Meta | Oracle | Other |
|---|---|---|---|---|---|---|
| Q1 2022 | $116.4B | $68.0B | $49.4B | $27.9B | $10.5B | $0.2B |
| Q2 2022 | $121.2B | $69.7B | $51.9B | $28.8B | $42.4B | $0.2B |
| Q3 2022 | $127.1B | $69.1B | $50.1B | $27.7B | $11.4B | $0.3B |
| Q4 2022 | $149.2B | $76.0B | $52.7B | $32.2B | $12.3B | $0.3B |
| Q1 2023 | $127.4B | $69.8B | $52.9B | $28.6B | $12.4B | $0.3B |
| Q2 2023 | $134.4B | $74.6B | $56.2B | $32.0B | $13.8B | $0.3B |
| Q3 2023 | $143.1B | $76.7B | $56.5B | $34.1B | $12.5B | $0.3B |
| Q4 2023 | $170.0B | $86.3B | $62.0B | $40.1B | $12.9B | $0.4B |
| Q1 2024 | $143.3B | $80.5B | $61.9B | $36.5B | $13.3B | $0.6B |
| Q2 2024 | $148.0B | $84.7B | $64.7B | $39.1B | $14.3B | $0.8B |
| Q3 2024 | $158.9B | $88.3B | $65.6B | $40.6B | $13.3B | $1.0B |
| Q4 2024 | $187.8B | $96.5B | $69.6B | $48.4B | $14.1B | $1.2B |
| Q1 2025 | $155.7B | $90.2B | $70.1B | $42.3B | $14.1B | $1.5B |
| Q2 2025 | $167.7B | $96.4B | $76.4B | $47.5B | $15.9B | $1.7B |
| Q3 2025 | $180.2B | $102.3B | $77.7B | $51.2B | $14.9B | $1.9B |
| Q4 2025 | $213.4B | $113.8B | $81.3B | $59.9B | $16.1B | $2.2B |
| Q1 2026 | $181.5B | $109.9B | $82.9B | $56.3B | $17.2B | $2.7B |
| Q2 2026 | $200.6B | $119.8B | $90.0B | $60.8B | $19.2B | $3.3B |
From spare cash to all of the cash
The most important number on this page is the ratio of capex to operating cash flow. It bottomed at 35% in Q3 2023 and reached 97% in Q2 2026. At a third, the build-out is an expense these companies fund from spare cash while still buying back stock. At nearly all, it competes with buybacks and dividends for every dollar, and any further growth must come from the balance sheet.
Company by company the margin of safety differs. Microsoft spent $136B against $196B of operating cash flow; Alphabet $160B against $180B. Amazon spent $173B against $161B, and Meta $104B against $133B. Oracle spent $75.7B against $46.9B: like Amazon, it spends more than its business generates, and by the widest margin, so it is the one to watch for the cost of funding.
Capex also understates the commitment. Microsoft, Meta, Oracle and Alphabet disclose $1.08T of data-center leases they have signed that haven’t started yet. Those payments will run for 15 to 20 years or more, and they don’t appear in capex or on the balance sheet until the buildings are ready. Following the money adds them up.
Capex by buyer, latest four quarters, and how much of operating cash flow it used
Cash paid for property and equipment
Source: company 10-K and 10-Q filings (SEC XBRL); fiscal quarters aligned to the nearest calendar quarter. Finance leases excluded.
Show the numbers
| Company | Capex | Operating cash flow | Capex ÷ cash flow |
|---|---|---|---|
| Amazon | $173B | $161B | 107% |
| Alphabet | $160B | $180B | 89% |
| Microsoft | $136B | $196B | 69% |
| Meta | $104B | $133B | 78% |
| Oracle | $75.7B | $46.9B | 161% |
| CoreWeave | $20.6B | $6.9B | 298% |
The return is the cloud businesses
If the spending pays, it pays through the cloud segments. AWS earned $148B of revenue over its latest four quarters, +28%, and $54.7B of operating income. Google Cloud earned $93.2B, +76%, and its operating income rose +198% to $31.8B. Microsoft Cloud, which also includes Microsoft 365 and Dynamics, brought in $229B. Oracle’s cloud infrastructure, the fastest-growing of them from a smaller base, reached $22.1B, +94%.
Two things in these numbers are encouraging. Growth has accelerated at every one of them over the past year, which is what you would expect if new capacity is being sold as it comes online. And operating income is growing at least as fast as revenue, which means the depreciation of the new equipment is being absorbed so far. The concern is the gap in scale: the clouds’ combined revenue is still well short of the buyers’ combined capex, and the depreciation of this year’s spending has barely started.
Backlog is the other evidence. Oracle reports $664 billion of remaining performance obligations and expects to recognize about 13% of it in the next twelve months. Microsoft reports $683B, Alphabet $568B. A backlog that large is demand made contractual, but it is only as good as the customers behind it, and several of the largest are AI labs funded by investors rather than by profits.
The GPU-only clouds
CoreWeave and Nebius are the purest listed bets on renting AI capacity. CoreWeave reported $7.6B of revenue over its latest four quarters, +115%, while spending $20.6B on equipment, 2.7× its revenue. Its backlog is $104B. The model is to sign long contracts with large customers and borrow against them to buy GPUs, so the business is as strong as its contracts and its access to credit.
Concentration is the risk to understand. CoreWeave reported that Microsoft accounted for about two-thirds of its 2025 revenue, and its largest new contracts are with OpenAI and Meta. Nvidia is both its sole GPU supplier and a shareholder. Nebius, which files a 20-F, reported $530M of revenue for 2025 and $4.1B of capex. These companies grow fastest when capacity is scarce; their test comes when it isn’t.
Cloudflare is a different kind of AI cloud. Rather than renting whole GPU clusters for training, it runs inference on GPUs spread across its global network, close to the users making requests. It reported $2.5B of revenue over its latest four quarters, +34%, at a 73% gross margin, while spending only $285M on capex. If more AI work shifts from training giant models to answering requests quickly, networks built for the edge are the part of this layer positioned for it.
What would change the picture
The bull case for this layer is simple: AI demand is real, the clouds are selling capacity as fast as they build it, and the businesses that own the most capacity will earn the most when demand matures, as they did in the first cloud cycle. The cloud segments’ accelerating growth and rising operating income support it.
The bear case is about returns on capital rather than demand. If capex keeps growing faster than cloud revenue, depreciation will compress margins even as revenue grows, and the stocks would be paying for growth that earns less than it used to. Watch for the moment capex guidance flattens. It will show whether the buyers think the capacity they have is enough, and it will hit every supplier layer below at once.
The companies, side by side
Latest four reported quarters for US filers; latest fiscal year for 20-F filers (marked). Underlined names have their own company page; the others link to the company’s filings on SEC EDGAR. Margins are left blank where a company doesn’t report gross profit or cost of revenue.
| Company | Revenue | vs year before | Gross margin | Operating margin | Capex | Backlog (RPO) |
|---|---|---|---|---|---|---|
| Amazon AMZNAWS; Anthropic's largest cloud partner and investor; Trainium chips | $776B | +16% | 51% | 12% | $173B | — |
| Alphabet GOOGLGoogle Cloud; TPUs designed in-house with Broadcom | $471B | +22% | 61% | 33% | $160B | $568B |
| Microsoft MSFTAzure; OpenAI's largest investor and a major cloud supplier to it | $346B | +18% | 68% | 47% | $136B | $683B |
| Meta METABuilds for its own apps; buys capacity from third-party clouds | $242B | +28% | 81% | 33% | $104B | — |
| Oracle ORCLOracle Cloud Infrastructure; the largest contracted backlog | $71.8B | +22% | — | 32% | $75.7B | $664B |
| CoreWeave CRWVGPU-only cloud; Microsoft, OpenAI and Meta as anchor customers | $7.6B | +115% | 67% | −3% | $20.6B | $104B |
| Cloudflare NETEdge network running AI inference close to users (Workers AI) | $2.5B | +34% | 73% | −14% | $285M | $2.7B |
| Nebius NBISGPU cloud built from Yandex's former international assetsFiscal year to December 31, 2025 (20-F) | $530M | — | — | −115% | $4.1B | — |
Revenue and growth. Total revenue over the four quarters, and its change from the four quarters before. Fast growth with a flat margin usually means volume; fast growth with a rising margin usually means pricing power.
Gross margin. The share of revenue left after the direct cost of what was sold. It is the best single measure of how much of each dollar a company keeps rather than passes on to its own suppliers.
Operating margin. What is left after research, sales and overhead too. The gap between gross and operating margin shows how much the company spends to stay competitive.
Capex. Cash paid for property and equipment over the four quarters: the spending that becomes revenue for the layers below.
Where the money comes from and goes
This layer is paid by software. Its own spending flows to servers, networking, power & DCs, chips. For the dollar amounts and who keeps them, see following the money.
Financing links that involve this layer
Quoted from the filings. Numbers match the money-loop map.
1Microsoft → OpenAIEquity investment
$11.9B funded of $13.0B committed
“We have made total funding commitments of $13.0 billion related to our investment, of which $11.9 billion has been funded as of June 30, 2026.”
Microsoft 10-K for the period ended June 30, 2026, filed July 29, 2026
2OpenAI → MicrosoftReported revenue
$24.1B of Microsoft revenue in FY2026
“For fiscal year 2026, we recorded revenue from commercial arrangements with OpenAI, inclusive of revenue-sharing payments, of $24.1 billion, and accounts receivable from OpenAI as of June 30, 2026 was $6.0 billion.”
Microsoft 10-K for the period ended June 30, 2026, filed July 29, 2026
3Amazon → AnthropicEquity investment
$8.0B in convertible notes, 2023–2025
“From Q3 2023 to Q4 2025, we invested $8.0 billion in convertible notes from Anthropic, which are classified as available-for-sale and reported at fair value with unrealized gains and losses included in "Accumulated other comprehensive income (loss)" and as Level 3 assets.”
Amazon 10-Q for the period ended June 30, 2026, filed July 31, 2026
4Amazon → AnthropicEquity investment
$5.0B in Series G preferred, 2026
“In Q2 2026, we invested $5.0 billion in Anthropic Series G nonvoting preferred stock.”
Amazon 10-Q for the period ended June 30, 2026, filed July 31, 2026
5Amazon → AnthropicCredit or guarantee
Up to $20B facility, released as compute is delivered
“Additionally, we entered into a financing arrangement to make available to Anthropic an aggregate facility not to exceed $20.0 billion that will expire 30 months after an Anthropic liquidity event, including an initial public offering ("IPO").”
Amazon 10-Q for the period ended June 30, 2026, filed July 31, 2026
6Anthropic → AmazonPurchase commitment
Commitment to AWS expanded by more than $100B over 10 years
“In Q2 2026, AWS and Anthropic announced an expansion of the strategic collaboration and existing multi-year commitment by more than $100.0 billion over 10.0 years, which includes contractual obligations related to the performance of AWS chips.”
Amazon 10-Q for the period ended June 30, 2026, filed July 31, 2026
7Amazon → OpenAIEquity investment
$50B of preferred stock: $15B invested, $35B committed
“We also invested $15.0 billion in Series C Preferred Stock of OpenAI and entered into an equity commitment letter agreement (the "Letter Agreement"), pursuant to which we agreed to purchase additional shares of Series C Preferred Stock (the "Commitment Shares") with an aggregate purchase price of $35.0 billion (the "Commitment Amount").”
Amazon 10-Q for the period ended June 30, 2026, filed July 31, 2026
8OpenAI → AmazonPurchase commitment
$38B AWS commitment, expanded by $100B over 8 years
“In Q1 2026, AWS and OpenAI Group PBC ("OpenAI") announced an expansion of the existing $38.0 billion multi-year commitment and commercial arrangement with OpenAI by $100.0 billion over 8.0 years, which includes contractual obligations related to the performance of AWS chips.”
Amazon 10-Q for the period ended June 30, 2026, filed July 31, 2026
9Amazon → AnthropicEquity investment
$5.0B in Series H preferred, drawn from the facility
“Under this financing arrangement, in Q2 2026, we exercised our option to participate in subsequent Anthropic equity financings by investing $5.0 billion in Anthropic Series H nonvoting preferred stock, which reduced the amount available under the facility to $15.0 billion.”
Amazon 10-Q for the period ended June 30, 2026, filed July 31, 2026
10OpenAI → CoreWeavePurchase commitment
Up to $11.9B through October 2030 (first agreement)
“For example, in March 2025, we entered into a master services agreement with OpenAI, a private company, pursuant to which OpenAI has committed to pay us up to approximately $11.9 billion through October 2030.”
CoreWeave 10-K for the period ended December 31, 2025, filed March 2, 2026
15Nvidia → CoreWeaveEquity investment
$2.0B share purchase, January 2026
“In January 2026, we entered into a securities purchase agreement with NVIDIA Corporation for a private placement of approximately 23 million shares of our Class A common stock at a purchase price of $87.20 per share, for aggregate gross proceeds of $2.0 billion.”
CoreWeave 10-Q for the period ended June 30, 2026, filed August 12, 2026
16Microsoft → CoreWeaveReported revenue
67% of CoreWeave’s 2025 revenue
“We recognized an aggregate of approximately 67% of our revenue from our top customer, Microsoft, for the year ended December 31, 2025.”
CoreWeave 10-K for the period ended December 31, 2025, filed March 2, 2026
17Meta → CoreWeavePurchase commitment
Up to $21.0B through 2032
“Similarly, in March 2026, we entered into an order form under an existing master services agreement pursuant to which Meta Platforms, Inc. ("Meta") initially committed to pay us up to approximately $21.0 billion (inclusive of access to new computing capacity through December 20, 2032 and the exercise of an existing option to access additional computing capacity through April 10, 2032) and, as a result, we also expect Meta to be a significant customer in future periods.”
CoreWeave 10-Q for the period ended June 30, 2026, filed August 12, 2026
18OpenAI → CoreWeavePurchase commitment
Up to $6.5B through May 2031 (one order form)
“In May 2025, we entered into a master services agreement with OpenAI OpCo, LLC ("OpenAI") and in September 2025, we entered into an order form under this master services agreement pursuant to which OpenAI committed to pay us up to approximately $6.5 billion through May 31, 2031 and, as a result, we expect OpenAI to be a significant customer in future periods.”
CoreWeave 10-Q for the period ended June 30, 2026, filed August 12, 2026
20AMD → MetaCustomer warrant
Warrant for up to 160M AMD shares at $0.01
“In October 2025 and February 2026, the Company issued warrants to OpenAI OpCo, LLC (OpenAI) and Meta Platforms, Inc. (Meta), each entitling the holder the right to purchase up to 160 million shares of the Company’s common stock at an exercise price of $0.01 per share (the OpenAI Warrant and the Meta Warrant, respectively).”
AMD 10-Q for the period ended June 27, 2026, filed August 5, 2026
21HPE → OracleCustomer warrant
Warrant for up to 4.2M HPE shares, tied to networking in Oracle’s AI data centers
“The Warrant was issued in connection with the Holder entering into a strategic collaboration with the Company to deploy the Company’s networking solutions in the Holder’s AI data centers.”
HPE 10-Q for the period ended July 31, 2026, filed September 3, 2026
22Meta → Meta’s data-center venturesCredit or guarantee
$41.5B of residual value guarantees, $68.4B maximum exposure
“The following table summarizes our maximum exposure to loss in the unconsolidated data center ventures (in millions): September 30, 2026 December 31, 2025 Carrying value of equity investments $4,732 $1,826 Lease commitments 18,476 12,311 Future funding commitments 3,703 3,724 Maximum residual value guarantees 41,493 28,091 Total $68,404 $45,952 Other In addition, we have other types of unconsolidated VIEs of which we are not the primary beneficiary.”
Meta 10-Q for the period ended September 30, 2026, filed October 22, 2026
23Meta → Third-party cloudsPurchase commitment
$454.5B of non-cancelable commitments, mostly cloud capacity
“As of September 30, 2026, we had $454.52 billion of non-cancelable contractual commitments, comprising both short-term and long-term arrangements, most of which are related to third-party cloud capacity arrangements and other investments in technical infrastructure, and we continue to enter into additional significant contractual arrangements.”
Meta 10-Q for the period ended September 30, 2026, filed October 22, 2026
24Microsoft → ConstellationPower contract
20-year PPA for a restarted nuclear plant
“The restart is supported by a 20-year PPA with Microsoft to purchase the output generated from the renewed plant.”
Constellation 10-Q for the period ended June 30, 2026, filed August 6, 2026
25Meta → VistraPower contract
2,609 MW of nuclear power under 20-year PPAs
“In January 2026, Vistra announced it had entered into 20-year PPAs with Meta, pursuant to which the Company has agreed to supply Meta with a total of 2,609 MW of carbon-free power and capacity from the Company's PJM nuclear power plants as follows: 1,268 MW of energy and capacity from Perry and 908 MW of energy and capacity from Davis-Besse; and 213 MW of uprate energy and capacity from Perry, 80 MW of uprate energy and capacity from Davis-Besse, and 140 MW of uprate energy and capacity from Beaver Valley.”
Vistra 10-Q for the period ended June 30, 2026, filed August 10, 2026
What to watch in the next filings
- Capex guidance on each quarterly call, and the capex-to-operating-cash-flow ratio in each 10-Q.
- Cloud segment growth and operating margin: AWS, Google Cloud, Microsoft’s Intelligent Cloud, Oracle’s cloud infrastructure.
- Changes in server useful lives. A shorter life means more depreciation and lower earnings.
- Debt issuance and credit spreads at Oracle and CoreWeave, the buyers furthest past self-funding.
- Leases signed but not yet started, and when they begin.
Free · quarterly · from SEC filings
Get the AI Stack Report every quarter
After each Big Tech earnings season we update the map from the new filings and publish what changed: who spent, who got paid, who kept it and what the filings reveal about the loop. Free, in Rob May’s Investing in AI newsletter.
Form not showing? Subscribe on Substack. No spam; unsubscribe any time.
Questions about AI clouds
Who are the biggest buyers of AI infrastructure?
Microsoft, Amazon, Alphabet, Meta and Oracle. Together they reported $566B of capital spending over their latest four quarters. GPU-focused clouds such as CoreWeave and Nebius are smaller but spend more relative to their revenue.
Can big tech afford its AI spending?
So far mostly from cash flow, but with less room than before. The five largest buyers spent 97% of their operating cash flow on capex in Q2 2026, against 35% at the low in Q3 2023. Oracle already spends more than it generates and funds the difference from its balance sheet.
How do the cloud companies earn a return on AI spending?
By renting the capacity to customers through their cloud businesses, and in Meta’s case by improving its own advertising products. Cloud segment revenue and operating income are the clearest reported evidence of the return.
What is CoreWeave?
A cloud company that rents GPU capacity, mostly built on Nvidia hardware and funded largely with debt secured by customer contracts. It reported $7.6B of revenue over its latest four quarters and a backlog of $104B. Microsoft was its largest customer in 2025.
Why does capex matter for AI stocks outside big tech?
Because the buyers’ capital spending is the revenue of the chip, memory, networking, server and power companies. When the buyers raise or cut their capex plans, every supplier layer feels it, usually within a quarter or two.
Research and education, not investment advice: nothing here is a recommendation to buy or sell any security. Figures come from companies’ SEC filings as reported; companies named were chosen because they show how the stack works, not as a view on their shares.